TRANSONIC REGULAR REFLECTION FOR THE NONLINEAR WAVE …
Transcript of TRANSONIC REGULAR REFLECTION FOR THE NONLINEAR WAVE …
April 3, 2006 12:49 WSPC/INSTRUCTION FILE nlws˙Jan9
Journal of Hyperbolic Differential Equationsc© World Scientific Publishing Company
TRANSONIC REGULAR REFLECTION FOR THE NONLINEAR
WAVE SYSTEM
KATARINA JEGDIC
Department of Mathematics, University of Houston
Houston, TX 77204-3008, USA
BARBARA LEE KEYFITZ∗
Fields Institute, 222 College Street
Toronto, Ontario M5T 3J1, Canada
SUNCICA CANIC
Department of Mathematics, University of Houston
Houston, TX 77204-3008, USA
Received (28 September 2005)Revised (10 January 2006)
Communicated by [editor]
Abstract. We consider Riemann data for the nonlinear wave system which result in aregular reflection with a subsonic state behind the reflected shock. The problem in self-similar coordinates leads to a system of mixed type and a free boundary value problemfor the reflected shock and the solution in the subsonic region. We show existence of asolution in a neighborhood of the reflection point.
Keywords: Two dimensional Riemann problems; nonlinear wave system; shock reflection.
1. Introduction
In this paper we continue the program initiated by Canic, Keyfitz, Kim and Lieber-
man on solving Riemann problems for two-dimensional systems of hyperbolic con-
servation laws modeling shock reflection. The first step in our approach is to write
the system in self-similar coordinates and obtain a system which changes type.
One finds a solution in the hyperbolic part of the domain using the standard the-
∗Permanent address: Department of Mathematics, University of Houston, Houston, TX 77204-3008, USA.
1
April 3, 2006 12:49 WSPC/INSTRUCTION FILE nlws˙Jan9
2 Katarina Jegdic, Barbara Lee Keyfitz, Suncica Canic
ory of one-dimensional hyperbolic conservation laws and the notion of quasi-one-
dimensional Riemann problems developed by Canic, Keyfitz and Kim (see [2] for
the unsteady small disturbance equation, [5] for the nonlinear wave system and
[3] for a general discussion). The position of the reflected shock is formulated as
a free boundary problem coupled to the subsonic state behind the shock through
the Rankine-Hugoniot conditions. To solve the free boundary problem behind the
reflected shock, one proceeds as follows: (1) fix a curve within a certain bounded set
of admissible curves approximating the free boundary, (2) solve the fixed boundary
problem, and (3) update the position of the reflected shock. This gives a mapping
on the set of admissible curves, and one proves there is a fixed point in a weighted
Holder space.
The idea was first implemented on a shock perturbation problem for the steady
transonic small disturbance equation by Canic, Keyfitz and Lieberman [8]. It was
extended to two types of regular reflection for the unsteady transonic small dis-
turbance equation in Canic, Keyfitz and Kim [4] (transonic regular reflection) and
Canic, Keyfitz and Kim [6] (supersonic regular reflection). The principal features
of this method for a class of two-dimensional conservation laws (including the un-
steady transonic small disturbance equation, the nonlinear wave system, and the
isentropic gas dynamics equations) are presented in the survey paper by Keyfitz
[16]. A detailed study of the subsonic solution to the fixed boundary problem for a
class of operators satisfying certain structural conditions is given in [15] by Jegdic,
Keyfitz and Canic.
In this paper we consider the two-dimensional nonlinear wave system (NLWS).
It can be obtained from the compressible isentropic gas dynamics equations by ne-
glecting the terms which are quadratic in the velocity, or by writing the nonlinear
wave equation as a first order system. Analogous to the gas dynamics equations,
when written in self-similar variables, the NLWS reduces to a second order quasi-
linear equation which changes type coupled to ”transport” equations, so that the
change of type takes one from the hyperbolic to a mixed type system. In the gas
dynamics equations, this coupling is highly nonlinear and leads to genuine technical
difficulties when analysis of such problems is attempted. However, in the NLWS, the
coupling between the second order quasilinear equation and the ”transport” equa-
tions is very weak and becomes important only when reconstructing the solution
in primitive variables. The NLWS is mathematically closely related to the pressure
gradient system studied by Zheng in [27]. More precisely, by assuming p(ρ) = eρ in
the NLWS and by replacing the dependant variable ρ by the pressure p, one gets
the pressure gradient system. It is sometimes also used as an approximation to the
shallow water equations, but then one usually takes the steady case (Pironneau,
[22]).
A partial solution to a Riemann problem for the NLWS leading to Mach reflec-
tion is given in [7] by Canic, Keyfitz and Kim. Solving a regular reflection problem
for this system extends the results of Canic, Keyfitz and Kim [4,6] to a more com-
plicated equation and boundary condition, and sets the stage for a further task (not
April 3, 2006 12:49 WSPC/INSTRUCTION FILE nlws˙Jan9
Transonic Regular Reflection for the Nonlinear Wave System 3
attempted in this paper), obtaining a global solution to a Riemann problem. We
also present an improved way of handling the artificial far-field Dirichlet boundary
condition. We take advantage of the simplified form of the NLWS in polar coordi-
nates. Change of variables to self-similar and polar coordinates is given in [5], as
well as the explicit solutions to quasi-one-dimensional Riemann problems that we
use here.
1.1. Related work
An overview of oblique shock wave reflection in steady, pseudo-steady and unsteady
flows from a phenomenological point of view is given in [1] by Ben-Dor. Existence
and stability of steady multidimensional transonic shocks was studied by Chen
and Feldman in [10]-[13]. Zheng [27] proved existence of a global solution to a
weak regular reflection for the pressure gradient system. Two-dimensional Riemann
problems for isentropic and polytropic gas dynamics equations were studied by
Zhang and Zheng in [24]. They give conjectures on the structure of the solutions
when initial data is posed in four quadrants and each jump results in exactly one
planar shock, rarefaction wave or slip plane far from origin. General mathematical
theory of two-dimensional Riemann problems for both scalar equations and systems
is presented by Zheng in [25]. An approach for proving existence of a global solution
to a weak regular reflection for polytropic gas dynamics equations with large gas
constant gamma is given in [26] by Zheng. Serre derives maximum principle for the
pressure and other a priori estimates in [23]. We also mention an earlier work of
Chang and Chen [9] on a formulation of a free boundary problem resulting from a
weak regular reflection for the polytropic gas dynamics equations.
1.2. Summary of the results
In Section 2 we state a Riemann problem for the NLWS resulting in a regular
reflection with a subsonic state behind the reflected shock. The discussion on how
the initial data is chosen so that the configuration leads to this type of reflection is
given in Appendix A. We write the problem in self-similar coordinates. Along the
lines of the study in [7], we find a solution in the hyperbolic part of the domain,
derive the equation of the reflected shock and give a formulation of the free boundary
problem behind the reflected shock. Our main result, Theorem 2.3, is local existence
of a solution to this free boundary problem, and the rest of the paper is devoted to
its proof.
In Section 3 we reformulate the problem using a second order elliptic equation
and from the Rankine-Hugoniot conditions along the free boundary we obtain an
oblique derivative boundary condition and an equation describing the position of
the reflected shock. To ensure that the problem is well-defined we introduce several
cut-off functions. This gives the modified free boundary problem stated in Theorem
3.1.
April 3, 2006 12:49 WSPC/INSTRUCTION FILE nlws˙Jan9
4 Katarina Jegdic, Barbara Lee Keyfitz, Suncica Canic
The first step in proving Theorem 3.1 is, as outlined above, to fix the position
of the free boundary within a bounded set of admissible curves and to solve the
modified fixed boundary problem. This task is completed in Section 4. We use the
study in [15] of fixed boundary value problems for a class of operators which satisfy
certain structural assumptions. For convenience we list those structural conditions
in the notation of this paper in Appendix B.
In Section 5 we use the Schauder fixed point theorem to show existence of a
solution to the modified free boundary problem.
Finally, the conditions under which a solution of the modified free boundary
problem solves the original free boundary problem are discussed in Section 6, com-
pleting the proof of Theorem 2.3.
2. The Statement of the Free Boundary Problem
In this section we formulate a Riemann problem leading to transonic regular re-
flection for the two-dimensional NLWS. The problem is considered in self-similar
coordinates, yielding a system which changes type. We find a solution in the hyper-
bolic part of the domain and formulate the problem for the position of the reflected
shock. The main result of the paper is local existence of a solution in the subsonic
part of the domain and is stated in Theorem 2.3.
The two-dimensional NLWS is a hyperbolic system of three conservation laws:
ρt +mx + ny = 0,
mt + px = 0, (t, x, y) ∈ [0,∞) × R × R. (2.1)
nt + py = 0,
Here, ρ : [0,∞)×R×R → (0,∞) stands for the density; m,n : [0,∞)×R×R → R
are “momenta” in the x and y directions, respectively; and p = p(ρ) is the pressure.
We denote c2(ρ) := p′(ρ), and we require that c2(ρ) be a positive and increasing
function for all ρ > 0.
We consider symmetric Riemann initial data (Fig. 1) consisting of two sectors
separated by the half lines x = ky and x = −ky, y ≥ 0, with k > 0. The data are
U(0, x, y) =
U0 = (ρ0, 0, n0), −ky < x < ky, y > 0,
U1 = (ρ1, 0, 0), otherwise,(2.2)
with the assumption ρ0 > ρ1 > 0. The constants k and n0 are specified in Section
2.1 (as described in Appendix A) in terms of ρ0 and ρ1 so that the Riemann problem
(2.1), (2.2) results in a regular reflection and we choose a solution (when there is
more than one) with a subsonic state behind the reflected shock.
Note that we can eliminate m and n in (2.1) and obtain a second order equation
for ρ alone:
ρtt = −mtx − nty = pxx + pyy = div(px, py) = div(c2(ρ)∇ρ), (2.3)
where “div” stands for the divergence and ∇ for the gradient in spatial variables.
We remark that (2.3) is a very natural nonlinearizing of the wave equation.
April 3, 2006 12:49 WSPC/INSTRUCTION FILE nlws˙Jan9
Transonic Regular Reflection for the Nonlinear Wave System 5
PSfrag replacements
x
y
x = kyx = −ky
U0
U1
Fig. 1. The Riemann initial data
We introduce self-similar coordinates ξ = x/t and η = y/t, and obtain
−ξρξ − ηρη +mξ + nη = 0,
−ξmξ − ηmη + pξ = 0,
−ξnξ − ηnη + pη = 0,
(2.4)
from (2.1), and the second order equation
((c2(ρ) − ξ2)ρξ − ξηρη)ξ + ((c2(ρ) − η2)ρη − ξηρξ)η + ξρξ + ηρη = 0, (2.5)
from equation (2.3). It is clear that when the equation (2.5) is linearized about a
constant state ρ > 0, the equation changes type across the sonic circle
Cρ : ξ2 + η2 = c2(ρ).
More precisely, (2.5) is hyperbolic outside of the circle Cρ and is elliptic inside.
2.1. Solution in the hyperbolic part of the domain
Suppose that the densities ρ0 > ρ1 > 0 are given. In this section we specify k and
n0, in terms of ρ0 and ρ1, so that a transonic regular reflection occurs, and we find
a solution to the Riemann problem (2.1), (2.2) in the hyperbolic region.
The parameter n0 = n0(ρ0, ρ1, k) is chosen so that each of the two discontinuities
x = ±ky, y ≥ 0, is resolved as a shock and a linear wave far from the origin (Fig.
2). From the calculation in [5, Appendix A] this means that given ρ0 > ρ1 > 0 and
k > 0, we take
n0 =
√1 + k2
k
√
(p(ρ0) − p(ρ1))(ρ0 − ρ1). (2.6)
Using the Rankine-Hugoniot relations, the one-dimensional Riemann solution
with states U0 on the left and U1 on the right consists of a linear wave la : ξ = kη,
April 3, 2006 12:49 WSPC/INSTRUCTION FILE nlws˙Jan9
6 Katarina Jegdic, Barbara Lee Keyfitz, Suncica Canic
PSfrag replacements
ξ
η
Ub Ua
U0
U1
Ξs
C0
C1
Cs
SaSb
lalb
Fig. 2. Interactions in the hyperbolic region
an intermediate state Ua = (ρ0,ma, na) and a shock Sa : ξ = kη + χa, with
χa = −√
1 + k2
√
[p]
[ρ], ma = −
√
[p][ρ]
1 + k2, na = −kma, (2.7)
where [ · ] denotes the jump between the states U0 and U1. By symmetry, the one-
dimensional solution in the left half-plane consists of a shock Sb : ξ = −kη − χa,
an intermediate state Ub = (ρ0,−ma, na) and a linear wave lb : ξ = −kη. Note that
the sonic circles for the states Ua and Ub coincide with the sonic circle for U0
C0 : ξ2 + η2 = c2(ρ0).
The first restriction on the choice of k = k(ρ0, ρ1) is that the point Ξs = (0, ηs)
where the shocks Sa and Sb meet should lie above the circle C0. We find
ηs =1
k
√
(1 + k2)(p(ρ1) − p(ρ0))
ρ1 − ρ0. (2.8)
Since the point Ξs is hyperbolic with respect to Ua and Ub, we solve a quasi-one-
dimensional Riemann problem at Ξs with states Ua and Ub, on the left and on
the right, respectively (with respect to an observer facing the origin) along a line
segment through Ξs which is parallel to the ξ-axis. A further restriction on the
value of k = k(ρ0, ρ1) is that this quasi-one-dimensional Riemann problem have a
solution (for details see Appendix A). In short, given ρ0 > ρ1 > 0, there exists a
value kC(ρ0, ρ1) with the property that if k is chosen so that
0 < k < kC , (2.9)
April 3, 2006 12:49 WSPC/INSTRUCTION FILE nlws˙Jan9
Transonic Regular Reflection for the Nonlinear Wave System 7
then the point Ξs is above the sonic circle C0 and, moreover, the quasi-one-
dimensional Riemann problem at Ξs with states Ua and Ub, on the left and on
the right, respectively, has a solution. From now on, we assume that the densities
ρ0 > ρ1 > 0 are fixed, that the parameter k is such that (2.9) holds and that the
momentum n0 is chosen as in (2.6).
Further, a calculation in Appendix A shows that if a solution to the above
quasi-one-dimensional Riemann problem at the reflection point Ξs exists, there
usually are two such solutions. Both consist of a shock connecting the state Ua to
an intermediate state and a shock connecting this intermediate state to Ub. Let us
denote the intermediate states for these two solutions by
UR = (ρR,mR, nR) and UF = (ρF ,mF , nF ).
More precisely (see Appendix A), we have
ρR, ρF > ρ0, mR = mF = 0, (2.10)
and we choose ρR < ρF . We find that c(ρF ) > ηs for all k ∈ (0, kC), and that
c(ρR) > ηs only when k is large enough, say k ∈ (k∗, kC), for some value k∗(ρ0, ρ1).
Therefore, the reflection point Ξs is subsonic with respect to the state UF for all
k ∈ (0, kC), and Ξs is subsonic with respect to the state UR if k ∈ (k∗, kC). We
denote the value of our solution at the reflection point Ξs by Us = (ρs,ms, ns), and
we choose
Us := U(Ξs) =
UR or UF , k ∈ (k∗, kC),
UF , k ∈ (0, k∗].(2.11)
This implies that the point Ξs is inside the sonic circle
Cs : ξ2 + η2 = c2(ρs).
As a consequence, the reflected shocks we study here are transonic throughout their
length. By causality, they cannot exit the sonic circle Cs and, by the Lax admis-
sibility condition (see [2] for the equivalent discussion on the unsteady transonic
small disturbance equation), they also do not cross the sonic circle C0 (Fig. 3).
Remark 2.1. If the two reflected shocks at the point Ξs were rectilinear, by the
Rankine-Hugoniot relations, their equations would be
η = ηs ± ξ
√
η2s
(p(ρs) − p(ρ0))/(ρs − ρ0)− 1.
2.2. Position of the reflected shock
Since the Riemann problem presented above is symmetric with respect to the η-
axis, from now till the end of the paper we restrict our attention to the right half
plane (ξ, η) : ξ ≥ 0. Writing (2.4) in polar coordinates
r =√
ξ2 + η2 and θ = arctan(η/ξ),
April 3, 2006 12:49 WSPC/INSTRUCTION FILE nlws˙Jan9
8 Katarina Jegdic, Barbara Lee Keyfitz, Suncica CanicPSfrag replacements
ξ
ηUb
Ua
U0
U1
Ξs
C0
C1
Cs
U1
U0
UaUb
U
Ξ0
SaSblalb
S
Density ( ), contour spacing = 0.15
η
−6 −4 −2 0 2 4 6 8−4
−2
0
2
4
6PSfrag replacements
ξ
η
Ub
Ua
U0
U1
Ξs
C0
C1
Cs
U1
U0
Ua
Ub
U
Ξ0
Sa
Sb
lalb
S
y−momentum (n), contour spacing = 0.75
η
−6 −4 −2 0 2 4 6 8−4
−2
0
2
4
6PSfrag replacements
ξ
η
Ub
Ua
U0
U1
Ξs
C0
C1
Cs
U1
U0
Ua
Ub
UΞ0
Sa
Sb
lalb
S
Fig. 3. Transonic regular reflection for the NLWS: definition of the states (top) and numericalsimulation showing the contour plot of ρ (bottom left) and the contour plot of n (bottom right).The inner circle on the bottom figures corresponds to the sonic circle C0, and the curve followingthe reflected wave corresponds to the numerically calculated transition between supersonic andsubsonic flow.
we obtain
−r cos θ sin θ
c2(ρ) cos θ −r 0
c2(ρ) sin θ 0 −r
Ur +1
r
0 − sin θ cos θ
−c2(ρ) sin θ 0 0
c2(ρ) cos θ 0 0
Uθ = 0,
April 3, 2006 12:49 WSPC/INSTRUCTION FILE nlws˙Jan9
Transonic Regular Reflection for the Nonlinear Wave System 9
or, in conservation form,
∂r
−rρ+m cos θ + n sin θ
p(ρ) cos θ − rm
p(ρ) sin θ − rn
+ ∂θ
1r (−m sin θ + n cos θ)
−p(ρ)r sin θ
p(ρ)r cos θ
=
−ρ− 1r (m cos θ + n sin θ)
−m− p(ρ)r cos θ
−n− p(ρ)r sin θ
.
Let S : r = r(θ), θ ∈ [−π/2, π/2], denote the reflected transonic shock in the right-
half plane. The Rankine-Hugoniot relations along S are
−r[ρ] + [m] cos θ + [n] sin θ =dr
dθ
1
r(−[m] sin θ + [n] cos θ)
[p] cos θ − r[m] = −[p]dr
dθ
sin θ
r
[p] sin θ − r[n] = [p]dr
dθ
cos θ
r, (2.12)
where U = (ρ,m, n) stands for the unknown solution behind the reflected shock
and [ · ] now denotes the jump between the states U0 and U . We express [m] and
[n] from the second and the third equations in (2.12), respectively, and substitute
into the first equation to obtain
dr
dθ= r
√
r2
s2− 1, (2.13)
with
s2 :=p(ρ0) − p(ρ)
ρ0 − ρ. (2.14)
Notice that this shock evolution equation is independent of m and n.
We recall the properties of s, from [7]
Lemma 2.2. Define the function
s(a, b) :=
√
p(a)−p(b)a−b , a, b > 0, b 6= a
c(a), b = a(2.15)
Then
(a) for fixed b > 0, the s(·, b) is increasing on (0,∞),
(b) limb→a s(a, b) = c(a), for a > 0, and
(c) if a > b > 0, then s(a, b) < c(a).
2.3. The statement of the main result
In this section we formulate the free boundary problem behind the reflected shock.
For the reasons explained in Section 3.2, we must exclude from our analysis the
point r(−π2 ) where the reflected shock intersects the η−axis (Ξ0 in Fig. 4). For
April 3, 2006 12:49 WSPC/INSTRUCTION FILE nlws˙Jan9
10 Katarina Jegdic, Barbara Lee Keyfitz, Suncica Canic
this reason, throughout the paper we fix an angle θ∗ ∈ (−π/2, π/2). We denote the
intersection of the reflected shock S and the line (r, θ∗) : r > 0 by V , and define
the closed line segment σ = [O, V ], where O is the origin; the vertical open line
segment Σ0 = (O,Ξs); and the open curve
Σ = (r(θ), θ) : θ ∈ (θ∗, π/2).The domain whose boundary is Ξs ∪ Σ ∪ σ ∪ Σ0 is denoted by Ω.
PSfrag replacements
ξ
η
Ub
Ua
U0
U1
Ξs
C0
C1
Cs
ρ1
ρ0
ρ
Ξ0
Sa
Sb
lalb
Σ
Σ0
σ
Ω
O
V
θ∗
Fig. 4. The domain Ω and its boundary.
We will impose a Dirichlet boundary condition for ρ along σ.
First, we define the set K of admissible shock curves. Suppose that ρ0 > ρ1 > 0
and k ∈ (0, kC(ρ0, ρ1)) are fixed. Let the parameter θ∗ ∈ (−π/2, π/2) be arbitrary.
We define the set K of candidate functions r(θ), θ ∈ [θ∗, π/2], describing the free
boundary Σ, by the following four properties.
• smoothness:
r(θ) ∈ H1+αK,
where αK ∈ (0, 1) will be chosen later and H1+αKis the Holder space
defined in Appendix C,
• conditions at the end point Ξs:
r(π/2) = ηs and r′(π/2) = ηs
√
η2s
s2(ρs, ρ0)− 1,
(the second condition comes from Remark 2.1)
April 3, 2006 12:49 WSPC/INSTRUCTION FILE nlws˙Jan9
Transonic Regular Reflection for the Nonlinear Wave System 11
• boundedness:
L ≤ r(θ) ≤ ηs, θ ∈ (θ∗, π/2),
• monotonicity:
L√
δ∗ ≤ r′(θ) ≤ ηs
√
η2s
c2(ρ0)− 1, θ ∈ (θ∗, π/2), (2.16)
where δ∗ > 0 will be specified later in terms of the fixed parameters ρ0, ρ1
and k.
A value of L we can use in this paper is
L :=ηs
e(π/2−θ∗)√
η2s/c2(ρ0)−1
.
We show this is an appropriate value in the proof of Lemma 5.1.
Although it is convenient to define the free boundary Σ by a curve r = r(θ)
in polar coordinates, we sometimes write Σ as ξ = ξ(η) in self-similar Cartesian
coordinates.
On σ we impose an artificial Dirichlet condition, ρ(r, θ∗) = f(r), chosen so that
ρ is larger than its value ρ0 outside Σ, and so that U is subsonic along σ. (These
are the properties that the global solution is expected to have along such a curve.)
Let ε∗ ∈ (0, ρs − ρ0) be fixed throughout the paper and let f : [0, ηs] → R be a
function in the Holder space Hγ;(0,ηs), for a parameter γ ∈ (0, 1) to be determined
later, such that
ρ0 + ε∗ ≤ f(r) ≤ ρs, c2(f(r)) > r2, 0 ≤ r ≤ ηs. (2.17)
With this notation we can now state the main result.
Theorem 2.3. (Free boundary problem )
Let the parameters ρ0 > ρ1 > 0 and k ∈ (0, kC(ρ0, ρ1)) be fixed. For every
θ∗ ∈ (−π/2, π/2) and ε∗ ∈ (0, ρs − ρ0), there exists γ0 > 0, depending on ρ0, ρ1, θ∗
and ε∗, such that for any γ ∈ (0,min1, γ0), αK = γ/2 and any function f ∈ Hγ
satisfying (2.17), the free boundary problem for ρ, m, n and r given by
−ξρξ − ηρη +mξ + nη = 0
−ξmξ − ηmη + pξ = 0
−ξnξ − ηnη + pη = 0
in Ω,
−r[ρ] + [m] cos θ + [n] sin θ = drdθ
1r (−[m] sin θ + [n] cos θ)
[p] cos θ − r[m] = −[p] drdθ
sin θr
[p] sin θ − r[n] = [p] drdθ
cos θr
on Σ,
r(π/2) = ηs,
ρ = f on σ, ρξ = 0 on Σ0, ρ(Ξs) = ρs,
April 3, 2006 12:49 WSPC/INSTRUCTION FILE nlws˙Jan9
12 Katarina Jegdic, Barbara Lee Keyfitz, Suncica Canic
has a solution ρ,m, n ∈ H(−γ)1+αK
and r ∈ H1+αKin a finite neighborhood of the
reflection point Ξs.
3. Derivation of the Modified Free Boundary Problem
Our main tool in proving Theorem 2.3 is the Holder theory of second order elliptic
equations, developed and expounded by Gilbarg, Trudinger and Lieberman. As
noted, we can reformulate the first order system in ρ, m and n (the subject of
Theorem 2.3) as a second order equation in ρ, (2.5), and in Section 3.1 we introduce a
cut-off function to keep the second-order equation strictly elliptic. Further, instead
of posing the Rankine-Hugoniot conditions (2.12) along the reflected shock, we
derive an oblique derivative boundary condition for ρ on Σ in Section 3.2. We
introduce a further cut-off function to ensure that the derivative boundary operator
on Σ is oblique. In Section 3.3 we modify the shock evolution equation (2.13) for
the reflected shock S, to ensure that it is well-defined. Thus, we obtain a problem
that does not involve m or n. (Towards the end of the paper, in Section 5, we show
how to recover m and n from the second and third equations in (2.4) by integrating
along the radial direction.) Finally, the modified free boundary problem is stated
in Section 3.4.
3.1. The second order operator for ρ
We recall the second order equation (2.5) for ρ, and we define the nonlinear operator
Q(ρ) := ((c2(ρ) − ξ2)ρξ − ξηρη)ξ + ((c2(ρ) − η2)ρη − ξηρξ)η + ξρξ + ηρη .
We rewrite (2.5) in polar coordinates and obtain
(
(c2(ρ) − r2) ρr
)
r+c2(ρ)
rρr +
(
c2(ρ)
r2ρθ
)
θ
= 0.
To ensure strict ellipticity of this equation, we introduce two cut-off functions
φi(x) :=
x, x > δiδi, x ≤ δi
i ∈ 1, 2, (3.1)
for δ1, δ2 > 0 to be determined in terms of the fixed parameters k, ρ0, ρ1 and ε∗.
The constant δ1 will be chosen in Section 6 and the constant δ2 will be specified in
(4.5). We modify each function φi so that it is smooth in a neighborhood of x = δi
and that φ′1 ∈ [0, 1] and φ′2 ≥ 0. We consider the modified equation
(
φ1(c2(ρ) − r2) ρr
)
r+c2(ρ)
rρr +
(
φ2(c2(ρ))
r2ρθ
)
θ
= 0. (3.2)
April 3, 2006 12:49 WSPC/INSTRUCTION FILE nlws˙Jan9
Transonic Regular Reflection for the Nonlinear Wave System 13
We rewrite equation (3.2) in self-similar Cartesian coordinates to get Q(ρ) = 0 with
Q(ρ) :=φ1 ξ
2 + φ2 η2
ξ2 + η2ρξξ + 2ξη
φ1 − φ2
ξ2 + η2ρξη +
φ1 η2 + φ2 ξ
2
ξ2 + η2ρηη
+
c2 − φ2
ξ2 + η2− 2φ′1
ξρξ + ηρη
+2cc′
ξ2 + η2
φ′1 (ξρξ + ηρη)2 + φ′2 (ηρξ − ξρη)2
, (3.3)
where the functions φ1 and φ′1 are evaluated at c2(ρ) − (ξ2 + η2), the functions φ2
and φ′2 are evaluated at c2(ρ), while c and c′ are evaluated at ρ. The eigenvalues of
the operator Q are
λ1(ρ) = φ1
(
c2(ρ) − (ξ2 + η2))
and λ2(ρ) = φ2
(
c2(ρ))
,
and Q is strictly elliptic since
λ(ρ) := minλ1(ρ), λ2(ρ) ≥ minδ1, δ2 > 0. (3.4)
3.2. Oblique derivative boundary condition
As in [7], we write system (2.4) in conservation form
∂ξ
m− ξρ
p− ξm
−ξn
+ ∂η
n− ηρ
−ηmp− ηn
= −2
ρ
m
n
.
The Rankine-Hugoniot relations along the reflected shock
S : ξ(η), η ≤ ηs, (3.5)
separating states U = (ρ,m, n) and U0 = (ρ0, 0, n0), are
[m]−ξ[ρ] =dξ
dη([n]−η[ρ]), [p]−ξ[m] = −dξ
dηη[m], −ξ[n] =
dξ
dη([p]−η[n]). (3.6)
As in [7], we derive the condition
β · ∇ρ = 0 on Σ. (3.7)
Here, ∇ρ := (ρξ , ρη) and β := (β1, β2) is given by
β1 = ξ′(ξ2 + η2)(c2(ρ) + s2)(ξ − ηξ′) − 2ξξ′s2(c2(ρ) + η2)
−2s2η(c2(ρ) − ξ2)(1 − (ξ′)2) + 2ξ′ξs2(ξ2 − c2(ρ)),
β2 = (c2(ρ) + s2)(ξ2 + η2)(ξ − ηξ′) + 2s2ξ′η(c2(ρ) − η2)
−2s2ξ(c2(ρ) − η2)(1 − (ξ′)2) + 2ξ′ηs2(c2(ρ) + ξ2),
(3.8)
where s2 is defined by (2.14). We define the operator
N(ρ) := β · ∇ρ. (3.9)
April 3, 2006 12:49 WSPC/INSTRUCTION FILE nlws˙Jan9
14 Katarina Jegdic, Barbara Lee Keyfitz, Suncica Canic
The operations by which the equations (2.13) and (3.7) are derived from the
Rankine-Hugoniot relations (2.12) and (3.6), respectively, can be reversed up to a
constant.
Let
ν :=1
1 + (ξ′)2(−1, ξ′)
denote the inward unit normal to the curve (3.5) describing the reflected shock, and
let us assume ξ(η) ≡ r(θ) ∈ K. We compute
β · ν =2s2(ξ′ξ + η)
1 + (ξ′)2
(c2(ρ) − η2)(ξ′)2 + 2ξηξ′ + c2(ρ) − ξ2
.
We remark that ξ′ξ + η = 0 if and only if the curve ξ(η) is tangent to a circle
centered at the origin, which is ruled out by the monotonicity property (2.16) of
the curves in the set K. (Note that by symmetry, the reflected shock S is tangent
to a circle at the point Ξ0, Fig. 4, and this is why we have to exclude Ξ0 from the
domain Ω.) Moreover, since the expression ξ′ξ+ η is positive at the reflection point
Ξs, the uniform monotonicity property of the curves in K implies that there exists
a constant C such that
ξ′ξ + η ≥ C > 0 (3.10)
holds uniformly in K. Further, we introduce the polynomial
P (Y ) := (c2(ρ) − η2)Y 2 + 2ξηY + c2(ρ) − ξ2, (3.11)
and remark that if P (ξ′) > 0, then β ·ν > 0 and the operatorN is oblique on Σ. Note
that P (ξ′(ηs)) > 0 and that the discriminant of P is negative if ξ2+η2 < c2(ρ(ξ, η)).
Thus, P (ξ′) > 0 holds at all points of the curve ξ(η) where ρ is strictly subsonic.
For the purpose of setting up an iteration, in which ρ may not always be subsonic
at every point on the curve ξ(η), we modify β by introducing a cut-off as follows.
We define a polynomial
G(Y ) :=
P (Y ) = (c2(ρ) − η2)Y 2 + 2ξηY + c2(ρ) − ξ2, ξ2 + η2 < c2(ρ) − δ1,
(ξY + η)2 + δ1(Y2 + 1), ξ2 + η2 ≥ c2(ρ) − δ1,
where δ1 is a positive parameter as in (3.1). We introduce a modification of β in
(3.8)
χ =
(β1, β2), ξ2 + η2 < c2(ρ) − δ1,
(χ1, χ2), ξ2 + η2 ≥ c2(ρ) − δ1,
(3.12)
in which c2 is replaced by ξ2 + η2 + δ1 when c2(ρ) ≤ ξ2 + η2 + δ1, so
χ1 = ξ′(ξ2 + η2)(ξ2 + η2 + δ1 + s2)(ξ − ηξ′) − 2ξξ′s2(ξ2 + 2η2 + δ1)
−2s2η(η2 + δ1)(1 − (ξ′)2) − 2ξ′ξs2(η2 + δ1),
χ2 = (ξ2 + η2 + δ1 + s2)(ξ2 + η2)(ξ − ηξ′) + 2s2ξ′η(ξ2 + δ1)
−2s2ξ(ξ2 + δ1)(1 − (ξ′)2) + 2ξ′ηs2(2ξ2 + η2 + δ1)
April 3, 2006 12:49 WSPC/INSTRUCTION FILE nlws˙Jan9
Transonic Regular Reflection for the Nonlinear Wave System 15
there. We define the operator
N(ρ) = χ · ∇ρ. (3.13)
Note that if ξ2 + η2 ≥ c2(ρ) − δ1, then
χ · ν =2s2(ξ′ξ + η)
1 + (ξ′)2
(ξ′ξ + η)2 + δ1((ξ′)2 + 1)
=2s2(ξ′ξ + η)
1 + (ξ′)2G(ξ′) ≥ 2s2(ξ′ξ + η)
1 + (ξ′)2δ1 > 0. (3.14)
Hence, if the boundary Σ is described by a curve ξ(η) ≡ r(θ) ∈ K, then the operator
N is uniformly oblique on Σ.
3.3. Shock evolution equation
In order for the equation of the reflected shock (2.13) to be well-defined we replace
it by
dr
dθ= r
√
ψ
(
r2
s2− 1
)
. (3.15)
Here
ψ(x) :=
x, x > δ∗δ∗, x ≤ δ∗,
(3.16)
where δ∗ is the same positive parameter as in (2.16) which will be specified in
Section 6 in terms of the a priori fixed parameters ρ0, ρ1 and k. Since we will need
ψ′ to be continuous, we modify ψ so that it is smooth in a neighborhood of x = δ∗.
3.4. The statement of the modified free boundary problem
Our objective is to prove existence of a solution to the following modified problem.
Theorem 3.1. (Modified free boundary problem )
Let ρ0 > ρ1 > 0, k ∈ (0, kC(ρ0, ρ1)), θ∗ ∈ (−π/2, π/2), ε∗ ∈ (0, ρs − ρ0) and
δ1 > 0 be given. There exist positive parameters δ∗, δ2 and γ0 such that for any
γ ∈ (0,min1, γ0), αK = γ/2 and any function f ∈ Hγ satisfying (2.17), the free
boundary problem for ρ and r given by
Q(ρ) = 0 in Ω,
N(ρ) = 0 on Σ,
r′(θ) = r√
ψ(
r2
s2 − 1)
on Σ, r(π/2) = ηs,
ρ = f on σ, ρξ = 0 on Σ0, ρ(Ξs) = ρs,
(3.17)
has a solution ρ ∈ H(−γ)1+αK
in Ω and r ∈ H1+αK.
We break the proof of Theorem 3.1 into two steps.
April 3, 2006 12:49 WSPC/INSTRUCTION FILE nlws˙Jan9
16 Katarina Jegdic, Barbara Lee Keyfitz, Suncica Canic
Step 1 is to solve the fixed boundary value problem obtained by replacing the
free boundary in Theorem 3.1 by a curve r chosen from the set K. Again, assume
we are given ρ0 > ρ1 > 0, k ∈ (0, kC(ρ0, ρ1)), θ∗ ∈ (−π/2, π/2), ε∗ ∈ (0, ρs − ρ0)
and the positive parameters δ1 and δ∗. We show that there exist δ2 > 0 and γ0 > 0,
depending only on ρ0, ρ1, k, θ∗, ε∗, δ1 and δ∗, such that for any γ ∈ (0,minγ0, 1),
αK ∈ (0,min1, 2γ), a fixed curve r ∈ K defining Σ and a function f ∈ Hγ
satisfying (2.17), the nonlinear fixed boundary problem
Q(ρ) = 0 in Ω,
N(ρ) = 0 on Σ,
ρ = f on σ, ρξ = 0 on Σ0, ρ(Ξs) = ρs,
(3.18)
has a solution ρ ∈ H(−γ)1+αK
in the domain Ω.
Step 2 is to define a mapping using the shock evolution equation. We update
the position of the reflected shock using the initial value problem
r′(θ) = r(θ)√
ψ( r(θ)2
s2(ρ(r(θ),θ),ρ0)− 1), θ ∈ (θ∗, π/2),
r (π/2) = ηs.(3.19)
This defines a map J : r 7→ r on the set K. We show that we can choose δ∗ in
terms of ρ0, ρ1 and k, so that there exists γ0 > 0 (possibly smaller than γ0 found
in the previous step), also depending on the a priori fixed parameters ρ0, ρ1, k, θ∗,
ε∗ and δ1, such that for any γ ∈ (0,min1, γ0) and αK = γ/2, the map J has a
fixed point r ∈ K. With this fixed point r(θ), θ ∈ (θ∗, π/2), defining the boundary
Σ = (r(θ), θ) : θ ∈ (θ∗, π/2), the corresponding solution ρ ∈ H(−γ)1+αK
to the fixed
boundary problem (3.18) solves the modified free boundary problem (3.17).
The first step is completed in Section 4 and the second in Section 5.
4. Solution to the Modified Fixed Boundary Problem
In this section we find positive parameters δ2 and γ0, depending only on ρ0, ρ1,
k, θ∗, ε∗, δ1 and δ∗, such that for any γ ∈ (0,minγ0, 1), αK ∈ (0,min1, 2γ), a
fixed r ∈ K describing the boundary Σ and a function f ∈ Hγ satisfying (2.17),
the fixed boundary problem (3.18) has a solution ρ ∈ H(−γ)1+αK
in Ω. We use the
result in Section 4 of [15] which applies to fixed nonlinear boundary problems of
the second order where the operators in the domain and on the boundary satisfy
certain structural conditions. These conditions are stated in Section 4.3 in [15] and,
for convenience, we give them in Appendix B using the notation of this paper.
We confirm in Proposition 4.1 that they hold for the problem (3.18), arising from
transonic regular reflection for the NLWS, and the result follows from Theorem 4.7
in [15].
Proposition 4.1. For any curve r ∈ K fixed, the boundary value problem (3.18)
satisfies the structural conditions (B.6)-(B.11). Moreover, for
K ≥ max
2c(ρs)c′(ρs)
δ1, 4(c′(ρs))
2
(4.1)
April 3, 2006 12:49 WSPC/INSTRUCTION FILE nlws˙Jan9
Transonic Regular Reflection for the Nonlinear Wave System 17
the inequality (B.12) holds.
Proof. First, we write the operator Q, given by (3.3), as in (B.5). We note that for
a fixed curve r ∈ K, the coefficients aij , bi and cij are in C1 and that the coefficients
χi of the vector χ, given by (3.12) and defining the operator N in (3.13) are such
that χi ∈ C2.
Recall that the operators Q and N , given by (3.3) and (3.13), are strictly elliptic
in Ω and oblique on Σ, respectively, by (3.4) and (3.14). Clearly, the operator
ρξ = (1, 0) · ∇ρ is both strictly and uniformly oblique on Σ0. Hence, the conditions
of Lemma 4.8 in [15] are satisfied. By this lemma, given r ∈ K, describing the
boundary Σ, and a solution ρ ∈ C1(Ω) to the fixed problem (3.18) we have uniform
L∞ bounds
ρ0 + ε∗ ≤ ρ(ξ, η) ≤ ρs, (ξ, η) ∈ Ω. (4.2)
Next, we show that the uniform bounds (4.2) imply uniform ellipticity of the
operator Q and both strict and uniform obliqueness of the operator N . The operator
Q is uniformly elliptic in Ω since
Λ(ρ)
λ(ρ):=
maxλ1(ρ), λ2(ρ)minλ1(ρ), λ2(ρ)
≤ c2(ρs)
minδ1, δ2. (4.3)
Further, recall the definition (3.12) of the vector χ and note that from (3.14), the
uniform bounds (4.2) on ρ and the uniform bounds on r ∈ K we have
χ · ν ≥ C > 0, for all ρ and r ∈ K,
for some constant C. Therefore, the operator N is strictly oblique on Σ. Moreover,
we have a uniform bound
|χ| =√
χ21 + χ2
2 ≤ C,
again using the uniform bounds on the curve ξ(η) ≡ r(θ) ∈ K describing the
boundary Σ and the uniform bound (4.2) on the solution ρ. Therefore
χ · ν|χ| =
2 s2 (ξ′ξ + η)G(ξ′)
|χ| ≥ 2 c2(ρ0 + ε∗) c δ1C
> 0, (4.4)
where c is the constant in (3.10). Hence, the operator N is also uniformly oblique
on Σ. This confirms that conditions (B.6)-(B.9) hold.
Note that by choosing δ2, in the definition (3.1) of the cut-off function φ2, such
that
0 < δ2 ≤ c2(ρ0 + ε∗), (4.5)
the function φ2 is equal to the identity function. We assume the choice (4.5) for δ2.
April 3, 2006 12:49 WSPC/INSTRUCTION FILE nlws˙Jan9
18 Katarina Jegdic, Barbara Lee Keyfitz, Suncica Canic
Therefore, the operator Q in (3.3) becomes
Q(ρ) =φ1 ξ
2 + c2 η2
ξ2 + η2ρξξ + 2ξη
φ1 − c2
ξ2 + η2ρξη +
φ1 η2 + c2 ξ2
ξ2 + η2ρηη
− 2φ′1 ξρξ + ηρη +2cc′
ξ2 + η2
φ′1 (ξρξ + ηρη)2 + 2cc′ (ηρξ − ξρη)2
,
=∑
i,j
aij(ρ, ξ, η)Dijρ+
∑
i
bi(ρ, ξ, η)Diρ+
∑
i,j
cij(ρ, ξ, η)DiρDjρ. (4.6)
Here, the functions c and c′ are evaluated at ρ, and φ1 and φ′1 are evaluated at
c2(ρ) − (ξ2 + η2).
Clearly, the condition (B.10) holds for the operator Q given by (4.6) and next
we check that (B.11) is also satisfied. We have
|∑
i,j
aij(ρ, ξ, η)Dijρ| ≤ (ηs + 2cc′ + 4c2(c′)2)(|ρξ |2 + |ρη|2) + 2ηs
≤ minδ1, δ2(
ηs + 2cc′ + 4c2(c′)2
minδ1, δ2∑
i
|Diρ|2 +2ηs
minδ1, δ2
)
.
Hence, (B.11) holds with
µ0 =ηs + 2c(ρs)c
′(ρs) + 4c2(ρs)(c′(ρs))
2
minδ1, δ2and Φ =
2ηs
minδ1, δ2. (4.7)
Finally, we check that (B.12) holds for the parameter K as in (4.1). Let r ∈ Kbe arbitrary and let ρ be a solution to the equation Q(ρ) = 0. It is easy to show
K∑
i,j
aij(ρ, ξ, η)DiρDjρ−
∑
i,j
cij(ρ, ξ, η)DiρDjρ =
1
ξ2 + η2
(Kφ1 − 2cc′φ′1)(ξρξ + ηρη)2 + (Kc2 − 4c2(c′)2)(ηρξ − ξρη)2
.
Hence, (B.12) holds.
Therefore, the structural conditions of Theorem 4.7 in [15] are satisfied. By this
theorem, there exists γ0 > 0, depending on the sizes of the opening angles of the
domain Ω at the set of corners V and on the bounds on the ellipticity ratio of the
operator Q, such that for every γ ∈ (0,minγ0, 1), αK ∈ (0,min1, 2γ), r ∈ Kand any function f ∈ Hγ satisfying (2.17), there exists a solution ρ to the fixed
boundary problem (3.18). Also, we have ρ ∈ H(−γ)1+α∗
, for all α∗ ∈ (0, αK].
Remark 4.2. We note that by the definition of the set K of admissible curves, the
sizes of the opening angles of the domain Ω at the set of corners V satisfy bounds
depending only on the parameters ρ0, ρ1, k and θ∗, which are fixed throughout the
paper, and on the parameter δ∗ which will be chosen in Section 5 also in terms of
ρ0, ρ1, k and θ∗. Therefore, the parameter γ0, given by Theorem 4.7 in [15], can be
taken independent of the choice of the curve r ∈ K. Moreover, using the uniform
April 3, 2006 12:49 WSPC/INSTRUCTION FILE nlws˙Jan9
Transonic Regular Reflection for the Nonlinear Wave System 19
bounds (4.3) on the ellipticity ratio of the operator Q and the choice of δ2 in (4.5),
we have that γ0 depends only on the fixed parameters ρ0, ρ1, k, θ∗ and ε∗, and the
parameters δ1 and δ∗ which will be chosen in Sections 5 and 6, respectively, also in
terms of ρ0, ρ1, k and ε∗.
5. Solution to the Modified Free Boundary Problem
In this section we complete the second step of the proof of Theorem 3.1.
Let γ0 > 0 be the parameter found in Section 4. Let γ ∈ (0,minγ0, 1) and let
αK ∈ (0,min1, 2γ) be arbitrary. For any r ∈ K, describing the boundary Σ, and
any function f ∈ Hγ satisfying (2.17), we find a solution ρ(ξ, η) to the nonlinear
fixed boundary problem (3.18). We define the curve r(θ), θ ∈ (θ∗, π/2), as a solution
to (3.19). This gives a map J : ρ 7→ ρ on the set K. We show that J has a fixed
point using the following
Theorem. (Corollary 11.2 in [14]) Let K be a closed and convex subset of a
Banach space B and let J : K → K be a continuous mapping so that J(K) is
precompact. Then J has a fixed point.
We take B to be the space H1+αK, and we take K as in Section 2.3. In this
section we specify the parameter δ∗ in the definition of the set K and the cut-off
function ψ (see (3.16)), and we further specify γ and αK so that the hypotheses of
the previous fixed point theorem are satisfied.
Lemma 5.1. Let the parameters ρ0 > ρ1 > 0, k ∈ (0, kC(ρ0, ρ1)), θ∗ ∈ (−π/2, π/2)
and ε∗ ∈ (0, ρs − ρ0) be given. Let δ∗ be such that
0 < δ∗ <η2
s
s2(ρs, ρ0)− 1. (5.1)
There exists γ0 > 0 such that for any γ ∈ (0,min1, γ0) and αK = γ/2, we have
(a) J(K) ⊆ K, and
(b) the set J(K) is precompact in H1+αK.
Remark 5.2. Recall from (2.10) that we have ρs > ρ0, implying, by the mono-
tonicity of the function s2(·, ρ0), that s2(ρs, ρ0) > c2(ρ0). Note that the choice of δ∗in (5.1) gives that
δ∗ < e(π−2θ∗)√
η2s/c2(ρ0)−1
(
η2s
c2(ρ0)− 1
)
,
and, in particular, the monotonicity condition (2.16) in the definition of the set Kmakes sense.
Proof. (of Lemma 5.1) This proof follows ideas from Section 4.2.1 in [4] and some
of its parts are identical to the proof of Lemma 5.3 in [15].
Let γ0 be the parameter found in Section 4. Let γ ∈ (0,minγ0, 1) be arbitrary
and let αK ∈ (0,min1, 2γ). Let r ∈ K and f ∈ Hγ satisfying (2.17) be given, and
April 3, 2006 12:49 WSPC/INSTRUCTION FILE nlws˙Jan9
20 Katarina Jegdic, Barbara Lee Keyfitz, Suncica Canic
let ρ(ξ, η) ∈ H(−γ)1+αK
be a solution to the fixed boundary problem (3.18) found in
Section 4. Further, suppose that r(θ), θ ∈ (θ∗, π/2), is a solution to the problem
(3.19).
To show (a) we need to show that ρ ∈ K. Clearly, r(π/2) = ηs, and
r′(π/2) = ηs
√
ψ
(
η2s
s2(ρs, ρ0)− 1
)
= ηs
√
η2s
s2(ρs, ρ0)− 1,
by the choice of δ∗. Next, note r′(θ) ≥ r(θ)√δ∗, implying dr
r ≥√δ∗. After integrat-
ing from θ to π/2, we get
r(θ) ≤ ηs, θ ∈ (θ∗, π/2). (5.2)
On the other hand,
r′(θ) ≤ r(θ)
√
ψ
(
η2s
c2(ρ0)− 1
)
≤ r(θ)
√
η2s
c2(ρ0)− 1, by the choice of δ∗,
implying drr ≤
√
η2s
c2(ρ0) − 1, and after integrating from θ to π/2 we obtain
r(θ) ≥ ηs
e(π/2−θ)√
η2s/c2(ρ0)−1
≥ ηs
e(π/2−θ∗)√
η2s/c2(ρ0)−1
.
Together with (5.2), this implies the desired boundedness of the curve r(θ). Once
this boundedness is established, the required monotonicity is clear.
It is left to show that we can find γ and αK so that
r ∈ H1+αK(5.3)
and that (b) holds. This part of the proof is identical to the proof of Lemma 5.3
in [15]. In short, Theorem 2.3 in [21] gives that there exist α0 and C such that a
solution ρ to the fixed boundary problem (3.18) satisfies
[ρ]α0≤ C
in a neighborhood of Σ. Here, α0 depends on the bounds for the ellipticity ratio of
the operator Q and on the obliqueness constant of the operator N , and on µ0|ρ|0,where µ0 is the constant in (4.7). The constant C also depends on Ω. Using the
bound (4.3) for the ellipticity ratio of Q and the choice (4.5) for δ2, the bound
(4.4) for the obliqueness constant of the operator N , uniform bounds (4.2) on the
solution ρ, the definition of the set K and the choice (5.1) for δ∗, we have that α0
and C depend only on the fixed parameters ρ0, ρ1, k, θ∗ and ε∗, and the parameter
δ1 which will be chosen in Section 6 also in terms of ρ0, ρ1, k and ε∗. We replace
γ0 by minγ0, α0 and we take γ ∈ (0,min1, γ0). This implies |r′|γ ≤ C and
|r|1+γ ≤ C(π/2 − θ∗). (5.4)
April 3, 2006 12:49 WSPC/INSTRUCTION FILE nlws˙Jan9
Transonic Regular Reflection for the Nonlinear Wave System 21
Therefore, r ∈ H1+γ . We choose αK ∈ (0, γ] to ensure (5.3). Since (5.4) holds
independently of r, we have that the set J(K) is contained in a bounded set in
H1+γ and to show (b) we take αK = γ/2.
We also note that the map J : K → K is continuous. Therefore, the hypothesis
of the fixed point theorem from the beginning of this section (Corollary 11.2 in
[14]) are satisfied and the map J has a fixed point r ∈ K. We use this curve r(θ),
θ ∈ (θ∗, π/2), to specify the boundary Σ, and using Section 4 we find a solution
ρ ∈ H(−γ)1+α∗
, for all α∗ ∈ (0, αK], of the modified free boundary problem (3.17).
Remark 5.3. Once the density component ρ is determined in the domain Ω, we
find the momenta m and n in Ω from the second and the third equations in (2.4).
These two equations are the transport equations for m and n:
∂m
∂s= pξ and
∂n
∂s= pη, (5.5)
where s = (ξ2 + η2)/2 stands for the radial variable. Note that m and n are known
in the hyperbolic part of the domain and along the boundary Σ using the Rankine-
Hugoniot relations (3.6). We find m and n in the domain Ω by integrating the
equations (5.5) from Σ towards the origin. We note that ∇p ∈ Hα and, hence, ∇pis absolutely integrable on Σ.
6. Proof of Theorem 2.3
In this section we discuss the conditions under which ρ, a solution to the modified
free boundary problem in Theorem 3.1, together with m and n as in Remark 5.3,
solves the free boundary problem in Theorem 2.3. More precisely, we investigate
when the cut-off functions φ1, φ2, χ and ψ can be removed. Recall that the functions
φ1 and φ2 are introduced in (3.1) so that the operator Q given by (3.3) is strictly
elliptic, the function χ is given by (3.12) and ensures that the operator N defined
in (3.13) is oblique and ψ, given by (3.16), is introduced so that the equation (3.15)
of the evolution of the reflected shock is well-defined.
Recall that we choose the parameter δ2 in the definition of φ2 so that the bounds
(4.5) hold. This implies that the cut-off function φ2 is identity.
Next we show that in a neighborhood of the reflection point Ξs = (0, ηs) the
cut-off functions φ1 and ψ can be replaced by identity and the cut-off function χ
can be replaced by β. Note that at Ξs we have
c2(ρ) − (ξ2 + η2) = c2(ρs) − η2s > 0,
because of our assumption that the point Ξs is subsonic with respect to the state
Us = (ρs,ms, ns) (see (2.11)). Further, note that at Ξs we have
r2
s2(ρ, ρ0)− 1 =
η2s
s2(ρs, ρ0)− 1 > 0 (6.1)
April 3, 2006 12:49 WSPC/INSTRUCTION FILE nlws˙Jan9
22 Katarina Jegdic, Barbara Lee Keyfitz, Suncica Canic
by Remark 2.1. Since the functions
c2(ρ) − (ξ2 + η2) andξ2 + η2
s2(ρ, ρ0)− 1, (ξ, η) ∈ Ω,
are positive at the reflection point Ξs, by continuity we have that these two functions
are positive in a closed neighborhood N of Ξs. We take the parameters δ1 and δ∗such that
δ1, δ∗ ∈(
0, min(ξ,η)∈N
c2(ρ) − (ξ2 + η2),ξ2 + η2
s2(ρ, ρ0)− 1
)
.
Hence, we can remove the cut-off functions φ1, ψ and χ in the neighborhood Nof the reflection point Ξs. Therefore, a solution ρ of the modified free boundary
problem in Theorem 3.1, with m and n found as in Remark 5.3, solves the free
boundary problem in Theorem 2.3 in the neighborhood N .
Acknowledgments
Much of this research was done during visits of the first author to the Fields In-
stitute, whose hospitality is acknowledged. We thank Gary Lieberman for useful
advice, and Allen Tesdall for contributing the numerical simulation shown in Fig.
3. Research of the first two authors was partially supported by the Department
of Energy, Grant DE-FG02-03ER25575, and by an NSERC Grant. Research of
the third author was partially supported by the National Science and Foundation,
Grants NSF FRG DMS-0244343, NSF DMS-0225948 and NSF DMS-0245513. We
also thank the Focused Research Grant on Multi-dimensional Compressible Euler
Equations for its encouragement and support.
Appendix A. Parameter Values for Regular Reflection
Consider the Riemann initial data (2.2) consisting of two sectors with states U0 =
(ρ0, 0, n0) and U1 = (ρ1, 0, 0), separated by half-lines x = ±ky, y ≥ 0, with k
positive, as in Fig. 1. We choose ρ0 > ρ1 > 0 arbitrary and we take
n0 =
√1 + k2
k
√
(p(ρ0) − p(ρ1))(ρ0 − ρ1).
This implies that each of the two initial discontinuities x = ±ky, y ≥ 0, results in
a one-dimensional solution consisting of a shock and a linear wave (Fig. 2). In this
part of the paper we describe how to choose the parameter k, depending on the
densities ρ0 and ρ1, so that the above Riemann data leads to a transonic regular
reflection.
Remark. Most of our discussion will be for a general function of pressure p(ρ),
ρ > 0, with property that
c2(ρ) := p′(ρ), ρ > 0, is a positive and increasing function. (A.1)
April 3, 2006 12:49 WSPC/INSTRUCTION FILE nlws˙Jan9
Transonic Regular Reflection for the Nonlinear Wave System 23
We will give more details for the example of the γ-law pressure with γ = 2. We
recall that a γ-law pressure relation is given by
p(ρ) = ργ/γ, ρ > 0,
for some γ > 1. We have c2(ρ) = ργ−1, ρ > 0, and we note that the system (2.1)
admits a scaling
(x, y) 7→ ρ(γ−1)/21 (x′, y′), ρ 7→ ρ1ρ
′ and (m,n) 7→ ρ(γ+1)/21 (m′, n′).
Hence, in this case, the flow behavior depends only on the density ratio ρ0/ρ1, or,
equivalently, on the velocity ratio or Mach number
M =c(ρ0)
c(ρ1)=
(
ρ0
ρ1
)(γ−1)/2
. (A.2)
Therefore, the Riemann data (2.2) can be parameterized in terms of ρ0/ρ1 and k.
Following the notation in Section 2.1, the one-dimensional Riemann solution
with states U0 and U1, on the left and on the right, respectively, consists of the
linear wave la : ξ = kη connecting U0 to the intermediate state Ua = (ρ0,ma, na)
and the shock Sa : ξ = kη + χa connecting Ua to U1. Further, the one-dimensional
solution with states U1 and U0, on the left and on the right, respectively consists
of the linear wave lb : ξ = −kη, the intermediate state Ub = (ρ0,−ma, na) and the
shock Sb : ξ = −kη − χa (see Fig. 2). Here, χa, ma and na are found using the
Rankine-Hugoniot relations and are given by (2.7).
Let Ξs = (0, ηs) denote the position of the projected intersection point of the
shocks Sa and Sb. Recall, that ηs is given by (2.8). We distinguish the following
three regions according to the position of the point Ξs:
• Region A: This region corresponds to those values of k, depending on ρ0
and ρ1, for which we have
ηs < c(ρ0). (A.3)
Hence, the point Ξs is inside the sonic circle C0 : ξ2 + η2 = c2(ρ0). The two
shocks Sa and Sb interact with C0 and a regular reflection cannot happen.
• Region B: In this region the parameter k(ρ0, ρ1) is specified so that we have
c(ρ0) < ηs < η∗,
where η∗ is the value below which the quasi-one-dimensional problem at
Ξs with states Ua and Ub on the left and on the right, respectively, does
not have a solution. Therefore, in this case, the shocks Sa and Sb could
intersect at the point Ξs, which is hyperbolic with respect to both states
Ua and Ub. However, a regular reflection cannot occur because the quasi-
one-dimensional problem at Ξs does not have a solution. We do not have
scenario for the solution in this region.
April 3, 2006 12:49 WSPC/INSTRUCTION FILE nlws˙Jan9
24 Katarina Jegdic, Barbara Lee Keyfitz, Suncica Canic
• Region C: The value of the parameter k(ρ0, ρ1) is such that
ηs > η∗. (A.4)
In other words, the shocks Sa and Sb intersect at the η-axis at the point
Ξs and moreover, the quasi-one-dimensional Riemann problem at Ξs has a
solution. Hence, a regular reflection occurs. We show in this section that
there are, in general, two solutions to this quasi-one-dimensional Riemann
problem, each consisting of two shocks. As in Section 2.1, we denote the
two intermediate states for these two solutions by
UR = (ρR,mR, nR) and UF = (ρF ,mF , nF ), (A.5)
where we assume that ρR ≤ ρF . We will further discuss for which values of
k satisfying (A.4) we have a transonic regular reflection and we will explain
the definition (2.11) of the state Us := U(Ξs).
The main goal of this section is to find the boundaries between the regions A,
B and C. Following the previous remark, in the case of a γ-law pressure, these
boundaries can be described by the curves in the (ρ0/ρ1, k)-plane. For γ = 2 they
are numerically computed in Fig. 5 (the region A being above the curve kA, the
region B is between kA and kC , and the region C is below the curve kC).
10 20 30 40 50 60Ρ0Ρ1
0.5
1
1.5
2
2.5
3k
PSfrag replacements
k = kC(ρ0/ρ1)
k = kA(ρ0/ρ1)
PSfrag replacements
k = kC(ρ0/ρ1)
k = kA(ρ0/ρ1)
Fig. 5. The curves kA and kC in the (ρ0/ρ1, k)-plane for the γ-law pressure, γ = 2.
Let us first consider the region A. Using the expression (2.8) for ηs, we get that
the condition (A.3) is equivalent to
k >s(ρ0, ρ1)
√
c2(ρ0) − s2(ρ0, ρ1)=: kA(ρ0, ρ1),
where s(·, ·) is defined in Lemma 2.2. Note that for a fixed ρ1 > 0 we have
limρ0→ρ1
kA(ρ0, ρ1) = ∞
April 3, 2006 12:49 WSPC/INSTRUCTION FILE nlws˙Jan9
Transonic Regular Reflection for the Nonlinear Wave System 25
and
limρ0→∞
kA(ρ0, ρ1) = limρ0→∞
√
p(ρ0)
ρ0c2(ρ0) − p(ρ0).
For a γ-law pressure we can write
kA
(
ρ0
ρ1
)
=
√
F (M)
γM2 − F (M), where F (M) =
(
M2γ
γ−1 − 1
M2
γ−1 − 1
)
,
and M is given by (A.2), and also
limρ0→∞
kA
(
ρ0
ρ1
)
=1√γ − 1
.
In the case γ = 2, we have (see Fig. 5)
kA
(
ρ0
ρ1
)
=
√
ρ0/ρ1 + 1
ρ0/ρ1 − 1.
Next we investigate the regions B and C, i.e., we suppose ηs > c(ρ0). Therefore,
the projections of the shocks Sa and Sb intersect at the point Ξs, hyperbolic with
respect to the states Ua = (ρ0,ma, na) and Ub = (ρ0,−ma, na), with values of ma
and na given in (2.7). We want to solve the quasi-one-dimensional Riemann problem
at Ξs, along a line segment parallel to the ξ-axis, with states Ub and Ua on the
left and on the right, respectively. (A general discussion on quasi-one-dimensional
Riemann problems is given in [3] and formulas for a solution in the case of the
NLWS are given in [5].) The condition for a solution to this quasi-one-dimensional
Riemann problem to exist is that the shock loci S+(Ua) and S−(Ub) intersect. The
formulas for m(ρ) along the shock loci S±(U), for a given state U , are obtained in
[5] (Appendix 6B). We have that if U = (ρ,m, n) ∈ S+(Ua), then
m(ρ) = ma +p(ρ) − p(ρ0)
ηs
√
η2s (ρ− ρ0)
p(ρ) − p(ρ0)− 1, (A.6)
and if U = (ρ,m, n) ∈ S−(Ub), then
m(ρ) = −ma − p(ρ) − p(ρ0)
ηs
√
η2s (ρ− ρ0)
p(ρ) − p(ρ0)− 1.
Moreover, along both shock polars S+(Ua) and S−(Ub) we have
n(ρ) = na +p(ρ) − p(ρ0)
ηs. (A.7)
Note that (A.7) implies that the intersections of the projected shock loci S+(Ua) and
S−(Ub) in the (ρ,m)-plane correspond to intersections of the loci in the (ρ,m, n)-
space.
In Figs. 6 and 7, we consider the γ-law pressure with γ = 2 and, for an example,
we take ρ0 = 64 and ρ1 = 1. For the case of k = 0.5, we plot the projected shock loci
April 3, 2006 12:49 WSPC/INSTRUCTION FILE nlws˙Jan9
26 Katarina Jegdic, Barbara Lee Keyfitz, Suncica Canic
50 100 150 200 250Ρ
-1000
-500
500
1000m
PSfrag replacements
S−(Ua)
S−(Ub)
S+(Ua)
S+(Ub)
PSfrag replacements
S−(Ua)
S−(Ub)
S+(Ua)
S+(Ub)
Fig. 6. The projected shock loci for the states Ua and Ub in the (ρ, m)-plane.
100 200 300 400Ρ
-1000
-500
500
1000
m
PSfrag replacements
k = 0.4
k = 0.5
k = 0.6053705
k = 0.7
25 50 75 100 125 150 175Ρ
-600
-400
-200
200
400
600
m
PSfrag replacements
k = 0.4
k = 0.5
k = 0.6053705
k = 0.7
PSfrag replacementsk = 0.4
k = 0.5
k = 0.6053705
k = 0.7
Fig. 7. The projected shock loci S+(Ua) and S−(Ub) in the (ρ, m)-plane for different values of k.
S±(Ua) and S±(Ub) in the (ρ,m)-plane in Fig. 6. In Fig. 7, we vary the parameter
k and depict the projections of the corresponding shock loci S+(Ua) and S−(Ub).
Note that because of the symmetry of the states Ua and Ub, the boundary
between the regions B and C occurs at those values of the parameter k = kC(ρ0, ρ1)
for which
max(ρ,m,n)∈S+(Ua)
m(ρ) = 0.
To find the values of kC(ρ0, ρ1) at which the maximum of the function m(ρ) given
by (A.6) is zero, we solve the system of equations m′(ρ) = 0 and m(ρ) = 0, i.e.,
c2(ρ)(ρ− ρ0) + p(ρ) − p(ρ0) − 2(p(ρ)−p(ρ0))c2(ρ)η2
s= 0,
m2a = (p(ρ) − p(ρ0))(ρ − ρ0) −
(
p(ρ)−p(ρ0)ηs
)2
.(A.8)
We express ρ from the first equation in (A.8) and substitute into the second equation
to find kC(ρ0, ρ1). Even in the case of the γ-law pressure with γ = 2, we obtain
only an implicit relation between kC and ρ0/ρ1, and we depict k = kC(ρ0/ρ1)
numerically in Fig. 5.
April 3, 2006 12:49 WSPC/INSTRUCTION FILE nlws˙Jan9
Transonic Regular Reflection for the Nonlinear Wave System 27
When k = kC(ρ0, ρ1), the two loci S+(Ua) and S−(Ub) are tangent, and they
intersect at a single point. This implies that there exists a unique solution to the
above quasi-one-dimensional Riemann problem at Ξs. When k is such that 0 < k <
kC(ρ0, ρ1), there are two points of intersection,
UR = (ρR,mR, nR) and UF = (ρF ,mF , nF ),
corresponding to different intermediate states in the two solutions of the quasi-one-
dimensional Riemann problem at the reflection point Ξs. Note that, by the geometry
of the shock loci S+(Ua) and S−(Ub) (see Fig. 6) and by symmetry, we have
ρR, ρF > ρ0 and mR = mF = 0.
We assume ρF > ρR. Each of the two solutions of the quasi-one-dimensional Rie-
mann problem consists of a shock connecting the state Ub to an intermediate state
(either UR or UF ) and a shock connecting this intermediate state to Ua.
For the case of the γ-law pressure with γ = 2, we find numerically that c(ρF ) =√ρF > ηs for any 0 < k < kC , and that c(ρR) =
√ρR > ηs, for sufficiently large
values of k. More precisely, the point Ξs is within the sonic circle
CR : ξ2 + η2 = c2(ρR)
only if k∗ < k < kC , for some value k∗(ρ0/ρ1). The curve k = k∗(ρ0/ρ1) is depicted
in Fig. 8. Hence, the reflection point Ξs is subsonic for the state UF for any 0 <
k < kC and is subsonic for the state UR if k∗ < k < kC , explaining our definition of
Us := U(Ξs) in (2.11). Even though we show this numerically for the γ-law pressure
with γ = 2 (we checked it also for γ = 3), we believe that it is true for any function
p(ρ) satisfying (A.1).
10 20 30 40 50 60Ρ0Ρ1
0.5
1
1.5
2
2.5
3k
PSfrag replacements
k = kA(ρ0/ρ1)
k = kC(ρ0/ρ1)
k = k∗(ρ0/ρ1)
k = kF (kC(ρ0/ρ1))
PSfrag replacementsk = kA(ρ0/ρ1)
k = kC(ρ0/ρ1)
k = k∗(ρ0/ρ1)k = kF (kC(ρ0/ρ1))
Fig. 8. The curves kA, kC and k∗ for the γ-law pressure, γ = 2.
By analogy with the gas dynamics equations or the unsteady transonic small
April 3, 2006 12:49 WSPC/INSTRUCTION FILE nlws˙Jan9
28 Katarina Jegdic, Barbara Lee Keyfitz, Suncica Canic
disturbance equation, we can think of UR and UF as “weak” and “strong” regular
reflection, respectively.
Appendix B. Structural Conditions for the Fixed Boundary
Problem
We state the structural conditions of Section 4.3 [15] in the notation of this paper.
First, we define
f(ξ, η) :=
f(ξ, η), (ξ, η) ∈ σ,
ρs, (ξ, η) = Ξs.(B.1)
By (2.17) we have
ρ0 + ε∗ ≤ f ≤ ρs,
which implies bounds on f independent of r ∈ K and of ρ. (This is the first condition
in (4.8) [15].) Note that, in the context of the NLWS, the second condition in (4.8)
[15] on the function f is replaced by
c2(f(r)) > r2 on σ = (r, θ∗) : 0 ≤ r ≤ r(θ∗),
ensuring that the solution is subsonic (as we have stated in (2.17)).
Next, we introduce the boundary operator N on Σ ∪ Σ0 as
N(ρ) := χ · ∇ρ, (B.2)
where the vector χ is defined by
χ :=
χ, on Σ,
(1, 0), on Σ0,(B.3)
with χ given by (3.12).
Then the fixed boundary value problem (3.18) can be written as
Q(ρ) = 0 in Ω,
N(ρ) = 0 on Σ := Σ ∪ Σ0,
ρ = f on ∂Ω \ Σ = σ ∪ Ξs,
(B.4)
where Q is the operator given in (3.3), N is given by (B.2) and f is given by (B.1).
Moreover, we write the operator Q as
Q(ρ) =∑
i,j
aij(ρ, ξ, η)Dijρ+
∑
i
bi(ρ, ξ, η)Diρ+
∑
i,j
cij(ρ, ξ, η)DiρDjρ. (B.5)
The structural conditions imposed on the problem (B.4) in [15] are as follows.
• The coefficients aij , bi and cij are in C1, and for a fixed curve r ∈ K we
have χi ∈ HαΣ.
April 3, 2006 12:49 WSPC/INSTRUCTION FILE nlws˙Jan9
Transonic Regular Reflection for the Nonlinear Wave System 29
• The operator Q is strictly elliptic, meaning
λ ≥ C1 > 0, for all ρ and r ∈ K, (B.6)
where λ denotes the smallest eigenvalue of the operator Q. We also assume
a bound on the ellipticity ratio of the form
Λ
λ≤ C2(|ρ|0), for all ρ and r ∈ K, (B.7)
where C2(|ρ|0) is a continuous function on R+. Here, Λ denotes the maxi-
mum eigenvalue of Q.
• The operator N is strictly oblique, i.e.,
χ · ν ≥ C3 > 0, for all ρ and r ∈ K, (B.8)
where ν stands for the unit inward normal to the boundary Σ. Also,
|χ| ≤ C4(|ρ|0), for all ρ and r ∈ K, (B.9)
holds, where C4(|ρ|0) is a continuous function on R+.
• For any solution ρ to the equation Q(ρ) = 0 in Ω we have
0 ≤∑
i,j
cij(ρ, ξ, η)DiρDjρ, (B.10)
and there exist µ0,Φ ∈ R, independent of ρ, such that
|∑
i,j
aij(ρ, ξ, η)Dijρ| ≤ λ
(
µ0
∑
i
|Diρ|2 + Φ
)
. (B.11)
It is noted in Remark 4.7 of [15] that, under the above conditions, a uniform
bound on the supremum norm |ρ|0, where ρ is any solution to the equation Q = 0
in Ω, implies the following:
• the norms |aij |0, |bi|0, |cij |0 and |χi|0 are uniformly bounded in ρ and r ∈ K,
and a uniform bound on the α-Holder seminorm [ρ]α implies that [aij ]α,
[bi]α, [cij ]α and [χi]α are uniformly bounded in ρ and r ∈ K (here, α ∈ (0, 1)
is arbitrary),
• the operator Q is uniformly elliptic,
• the boundary operator N is uniformly oblique, and
• since the matrix [aij(ρ, ξ, η)] is uniformly positive definite and the coeffi-
cients cij(ρ, ξ, η) are uniformly bounded, there exists K > 0, independent
of ρ and r ∈ K, such that∑
i,j
cij(ρ, ξ, η)DiρDjρ ≤ K
∑
i,j
aij(ρ, ξ, η)DiρDjρ. (B.12)
(This constant K plays role in the construction of a subsolution to the non-
linear fixed boundary problem (B.4) which is used to show that a solution
to (B.4) exists. For more details, see Lemma 4.14 in [15]).
April 3, 2006 12:49 WSPC/INSTRUCTION FILE nlws˙Jan9
30 Katarina Jegdic, Barbara Lee Keyfitz, Suncica Canic
In [15] the additional condition Σ ⊂ H2+α, where α ∈ (0, 1), was imposed. However
in steps 3 and 4 of the proof of Theorem 4.11 we showed how to eliminate this
condition. This, we require only the weaker hypothesis Σ ∈ H1+αΣleading to χ ∈
HαΣ.
Appendix C. Definitions of Weighted Holder Spaces
For a set S ⊆ R2 and a function u : S → R, we recall the definitions of the following
seminorms and norms (for more details see [14]):
|u|0;S := supX∈S |u(X)| supremum norm,
[u]α;S := supX 6=Y|u(X)−u(Y )|
|X−Y |α α-Holder seminorm,
|u|α;S := |u|0;S + [u]α;S α-Holder norm,
|u|k+α;S :=∑k
i=0 |Diu|0;S + [Dku]α;S (k + α)-Holder norm.
Here, α ∈ (0, 1), k is a nonnegative integer and Diu denotes the collection of the
i−th order derivatives of u.
In the definition of the set K and in Theorem 2.3, H1+α denotes the space of all
curves r(θ), θ ∈ (θ∗, π/2), such that
|r|1+α;(θ∗,π/2) <∞,
and H(−γ)1+α denotes the space of functions u(ξ, η), (ξ, η) ∈ Ω, such that
|u|(−γ)1+α;Ω\V := sup
δ>0δ1+α−γ |u|1+α;Ωδ;V
<∞,
with Ωδ;V := X ∈ Ω : dist(X,V) > δ and V := O, V,Ξs.
References
[1] G. Ben-Dor, Shock wave reflection phenomena (Springer-Verlag, New York, 1992).[2] S. Canic, B. L. Keyfitz, Riemann problems for the two-dimensional unsteady transonic
small disturbance equation, SIAM Journal on Applied Mathematics 58 (1998) 636–665.[3] S. Canic, B. L. Keyfitz, Quasi-one-dimensional Riemann problems and their role in
self-similar two-dimensional problems, Archive for Rational Mechanics and Analysis144 (1998) 233–258.
[4] S. Canic, B. L. Keyfitz, E. H. Kim, Free boundary problems for the unsteady transonicsmall disturbance equation: transonic regular reflection, Methods of Application andAnalysis 7 (2000) 313–336.
[5] S. Canic, B. L. Keyfitz, E. H. Kim, Mixed hyperbolic-elliptic systems in self-similarflows, Boletim da Sociedade Brasileira de Matematica 32 (2001) 377–399.
[6] S. Canic, B. L. Keyfitz, E. H. Kim, A free boundary problem for a quasilinear degener-ate elliptic equation: Regular reflection of weak shocks, Communications on Pure andApplied Mathematics LV (2002) 71–92.
[7] S. Canic, B. L. Keyfitz, E. H. Kim, Free boundary problems for nonlinear wave systems:Mach stems for interacting shocks, SIAM Journal on Mathematical Analysis, to appear.
[8] S. Canic, B. L. Keyfitz, G. Lieberman, A proof of existence of perturbed steady tran-sonic shocks via a free boundary problem, Communications on Pure and Applied Math-ematics LIII (2000) 1–28.
April 3, 2006 12:49 WSPC/INSTRUCTION FILE nlws˙Jan9
Transonic Regular Reflection for the Nonlinear Wave System 31
[9] T. Chang, G. Q. Chen, Diffraction of planar shocks along compressive corner, ActaMathematica Scientia 3 (1986) 241–257.
[10] G. Q. Chen, M. Feldman, Multidimensional transonic shocks and free boundary prob-lems for nonlinear equations of mixed type, Journal of the American MathematicalSociety 16 (2003) 461–494.
[11] G. Q. Chen, M. Feldman, Steady transonic shocks and free boundary problems ininfinite cylinders for the Euler equations, Communications on Pure and Applied Math-ematics 57 (2004) 310–356.
[12] G. Q. Chen, M. Feldman, Free boundary problems and transonic shocks for the Eulerequations in unbounded domain, Ann. Sc. Norm. Super. Pisa C1. Sci. 5 (2004) 827–869.
[13] G. Q. Chen, M. Feldman, Existence and stability of multidimensional transonic flowsthrough an infinite nozzle of arbitrary cross sections, submitted.
[14] D. Gilbarg, N. S. Trudinger, Elliptic Partial Differential Equations of Second Order(Springer - Verlag, New York, 2nd edition, 1983).
[15] K. Jegdic, B. L. Keyfitz, S. Canic, Transonic regular reflection for the unsteady tran-sonic small disturbance equation - details of the subsonic solution, Proceedings of theIFIP Conference: Free and Moving Boundaries Analysis, Simulation and Control; JohnCagnol, Jean-Paul Zolesio (Eds.) (Houston, 2005).
[16] B. L. Keyfitz, Self-Similar Solutions of Two-Dimensional Conservation Laws, Journalof Hyperbolic Differential Equations 1 (2004) 445–492.
[17] G. M. Lieberman, The Perron process applied to oblique derivative problems, Ad-vances in Mathematics 55 (1985) 161–172.
[18] G. M. Lieberman, Mixed boundary value problems for elliptic and parabolic differ-ential equations of second order, Journal of Mathematical Analysis and Applications113 (1986) 422–440.
[19] G. M. Lieberman, Local estimates for subsolutions and supersolutions of obliquederivative problems for general second order elliptic equations, Transactions of theAmerican Mathematical Society 304 (1987) 343–353.
[20] G. M. Lieberman, Optimal Holder regularity for mixed boundary value problems,Journal of Mathematical Analysis and Applications 143 (1989) 572–586.
[21] G. M. Lieberman, N. S. Trudinger, Nonlinear oblique boundary value problems fornonlinear elliptic equations, Transactions of the American Mathematical Society 295
(1986) 509–546.[22] O. Pironneau, Private communication.[23] D. Serre, Shock reflection in gas dynamics, Handbook of Mathematical Fluid Dynam-
ics; Susan Friedlander, Denis Serre (Eds.) 4 (Elsevier, North-Holland, 2005).[24] T. Zhang, Y. Zheng, Conjecture on the structure of solutions of the Riemann problems
for two-dimensional gas dynamics systems, SIAM Journal on Mathematical Analysis21 (1990) 593–630.
[25] Y. Zheng, Systems of Conservation Laws: Two-dimensional Riemann Problems(Birkhauser, 2001).
[26] Y. Zheng, Shock reflection for the Euler system, Proceedings of the 10th InternationalConference on Hyperbolic Problems, Theory, Numerics, Applications (Osaka, 2004).
[27] Y. Zheng, Two-dimensional regular shock reflection for the pressure gradient systemof conservation laws, preprint.